Mobile Phone Coverage and Producer Markets: Evidence from West Africa Jenny C. Aker and Marcel Fafchamps* Mobile phone coverage has expanded considerably throughout the developing world, particularly within sub-Saharan Africa. Existing evidence suggests that increased access to information technology has improved agricultural market efficiency for con- sumer markets and certain commodities, but there is less evidence of its impact on pro- ducer markets. Building on the work of Aker (2010), we estimate the impact of mobile phone coverage on producer price dispersion for three commodities in Niger. Our results suggest that mobile phone coverage reduces spatial producer price dispersion by 6 percent for cowpea, a semi-perishable commodity. These effects are strongest for remote markets and during certain periods of the year. The introduction of mobile phone coverage has no effect on producer price dispersion for millet and sorghum, two staple grains that are less perishable and are commonly stored by farmers. There are no impacts of mobile phone coverage on producer price levels, but mobile phone cover- age is associated with a reduction in the intra-annual price variation for cowpea. JEL classification codes: JEL O1, O3, Q13 INTRODUCTION Price information plays an important role in arbitrage behavior and market efficiency. Access to such information has improved significantly in developed countries, especially with the introduction of online databases (Autor 2001; Anderson and Magruder 2012). In sub-Saharan Africa, long distances and limited infrastructure have historically made obtaining such information costly. The spread of mobile phone coverage over the past decade has significantly reduced the costs of obtaining such information, thereby enabling consumers, traders, and producers to send and receive information more quickly and cheaply * Jenny C. Aker (corresponding author): The Fletcher School and Department of Economics, Tufts University, 160 Packard Avenue, Medford, MA 02155. Email: Jenny.Aker@tufts.edu. Marcel Fafchamps: Freeman Spogli Institute for International Studies, Stanford University, 616 Serra Street, Stanford CA 94305. Email: fafchamp@stanford.edu. This research was partially funded by Rocca Dissertation Fellowship, the Ford Foundation, and UC Berkeley’s CIDER. We are grateful for comments from three anonymous referees, Reena Badiana, Joanna Upton, and participants at the Center for the Study of African Economies (CSAE) and American Economic Association (AEA) conferences. All errors are our own. A supplemental appendix to this article is available at http://wber.oxfordjournals.org/. THE WORLD BANK ECONOMIC REVIEW, VOL. 29, NO. 2, pp. 262– 292 doi:10.1093/wber/lhu006 Advance Access Publication October 8, 2014 # The Author 2014. Published by Oxford University Press on behalf of the International Bank for Reconstruction and Development / THE WORLD BANK. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com. 262 Aker and Fafchamps 263 (Aker and Mbiti 2010). Although a growing body of evidence suggests that this increased access to information has improved consumer market efficiency, there is more limited evidence on its impact on producer markets, which form an im- portant segment of the population, especially in developing countries. Between 2001 and 2008, mobile phone networks were phased in throughout Niger. By 2008, more than 44 percent of the population and 90 percent of agri- cultural markets in our sample had access to mobile phone coverage (Aker 2010; authors’ calculations from GSMA). As coverage remained relatively limited in rural areas by 2008, the introduction of mobile phones primarily reduced traders’ cost of obtaining market information (Aker 2010), and, to a lesser extent, farmers who transported their output to local markets for sale. In earlier work, Aker (2010) finds that the rollout of mobile phone coverage in Niger reduced millet consumer price dispersion by 10 –16 percent. This paper estimates the impact of mobile phone coverage on producer price dispersion for three agricultural commodities in Niger. We first develop a simple model showing how the circulation of information among traders is likely to improve spatial arbitrage and reduce price dispersion in producer markets. These effects should be strongest for perishable and semi-perishable commodities, when inter- temporal arbitrage is largely absent. However, the theoretical predictions related to producers’ prices and traders’ gross margins are ambiguous, as they depend on the market structure and relative elasticities of supply and demand for a particu- lar commodity. We test these theoretical predictions using market-level panel data on produc- er prices for millet, sorghum, and cowpea in Niger between 1999 and 2008. We find that mobile phone coverage reduces the spatial dispersion of producer prices by 6 percent for cowpea, a semi-perishable cash crop in Niger. These effects are larger for more isolated markets and during certain periods of the year. We find no effect of mobile phone coverage on producer price dispersion for millet or sorghum, or on producer price levels. Nevertheless, mobile phone coverage reduced intra-annual producer price variation for cowpea by 6 percent. We use farmer-level data to better understand the correlations between mobile phone coverage and farmers’ marketing behavior. While farmers were more likely to use information technology as a search tool, mobile phone ownership was not correlated with changes in farmers’ marketing behavior or the sales prices that they received, consistent with Fafchamps and Minten (2012) and Aker and Ksoll (2013). Yet Tack and Aker (2014) find evidence that mobile phone coverage in- creased traders’ search behavior over time, suggesting that most of the reductions in search costs might have accrued to traders during this period. Our paper makes two contributions. First, the results speak to a substantial economic literature showing that information is crucial for the effective function- ing of markets, both from a theoretical (Stigler 1961; Reinganum 1979; Stahl 1989) and an empirical perspective (Autor 2001; Brown and Goolsbee 2002; Jensen 2007; Aker 2010; Goyal 2010). Most of this work has focused on con- sumer markets, with a more recent extension to producer prices (Jensen 2007; 264 THE WORLD BANK ECONOMIC REVIEW Muto and Yamano 2009; Goyal 2010; Fafchamps and Minten 2012). However, few of these studies assess the impact of information technology on agricultural producer price dispersion, as our work does. Second, much of the existing literature concentrates on the impact of informa- tion technology on a single good (Jensen 2007; Aker 2010; Goyal 2010). To our knowledge, only Muto and Yamano (2009) and Nakasone (2013) assess the impact of information technology on perishable and non-perishable commodi- ties. Our study is able to measure the impact of information technology on multiple goods with varying perishability and over a long time period. The rest of this paper proceeds as follows. We first provide an overview of the context and research design. Next, we present the theoretical framework. Subsequently, we present our data and then discuss the empirical strategy. Finally, we provide the main empirical results and present conclusions. CONTEXT With an estimated 61 percent of the population living in poverty, Niger is one of the lowest-ranked countries on the United Nations Human Development Index (United Nations Development Program 2013). Agriculture employs more than 80 percent of the population and contributes approximately 43 percent to the GDP (Aker 2010; World Bank 2013). The majority of the population consists of subsistence farmers who depend on rain-fed agriculture and livestock as their main sources of income. Subsistence farmers primarily cultivate millet, sorghum, and cowpea with a unimodal rainfall system. Although these commodities can, in principle, be stored for several years, a majority of the farmers and traders in our study area store for shorter periods and rarely engage in inter-annual storage (Aker 2008; Tack and Aker 2014).1 Cowpea, in particular, is highly susceptible to storage insects (the cowpea weevil), with storage losses estimated to be 25 percent (Jackai and Daoust 1986). As a result, cowpea is often considered to be a semi- perishable commodity (Murdock et al. 1997). Although a majority of farmers in our sample are net consumers, they often sell a portion of their production at some point in the year. All three commodities are traded through a system of weekly agricultural markets (Aker 2010). Farmers typically sell their agricultural products to smaller traders (e.g., retailers and in- termediaries) located within their village or at a nearby weekly market (an average of 7.5 km away). These smaller traders sell to wholesalers and semi- wholesalers in local markets, who, in turn, sell to buyers (wholesalers, retailers, or consumers) in regional markets. Although an agricultural market information 1 Aker (2008) shows that grain traders store an average of one week, with larger traders (wholesalers) storing for an average of one month. Storage duration data were not available by commodity. Farmer-level data collected by one of the authors suggests that farmers primarily produce cowpea as a cash crop and sell soon after the harvest, thus implying limited storage, and store millet and sorghum for longer periods. Aker and Fafchamps 265 system has existed in Niger since the 1990s, 89 percent of traders and 75 percent of farmers stated that they primarily obtain price information through personal networks (Aker 2010).2 Mobile phone service was introduced in Niger in October 2001 (Aker 2010). Three private mobile phone operators intended to provide universal coverage by 2009, with mobile phone service rolled out in different markets over time. At the outset, mobile phone operators prioritized urban centers and a town’s proximity to an international border.3 As a result, the capital city and regional capitals received coverage during the first three years of mobile phone rollout, followed by a quasi-experimental pattern in later years. Aker (2010) shows the spatial rollout of mobile phone coverage by market and by year between 2001 and 2008. Mobile phone coverage increased substan- tially during this time, with 90 percent of the weekly markets having access to mobile phone service by 2008. The greatest increase in mobile service into more remote rural areas occurred between 2008 and 2010, after our study period. While landlines have existed in Niger for some time, Niger has the second- lowest landline coverage in the world (World Bank 2006), and the number of landlines remained relatively stable during our study period (Figure 1).4 Among all of the agricultural markets in our study, only one received new landline coverage between 1999 and 2008. Despite a large increase in mobile phone coverage, Niger had the lowest adop- tion rate in Africa in 2008. There were an estimated 1.7 million mobile phone subscribers in 2008, representing 12 percent of the population (authors’ calcula- tions from Wireless Intelligence). Yet grain traders, especially wholesalers, were some of the first adopters: A trader survey conducted in 2006 showed that ap- proximately 30 percent owned a mobile phone and used it for their trading oper- ations (Aker 2010; Tack and Aker 2014). In contrast, less than 5 percent of farm households owned a mobile phone at that time.5 CONCEPTUAL FRAMEWORK To clarify the way in which we expect mobile phone coverage to affect producer price dispersion, we present a model of agricultural markets visited by 2 In surveys with agricultural traders and producers between 2005 and 2007, an overwhelming majority stated that they did not access or use price information provided Agricultural Market Information System (AMIS), primarily because of the type of data (only consumer prices are provided) and the timing of the data diffusion (the data are provided weekly, in some cases six days after a market). 3 Based on one of the author’s interviews with mobile phone service providers in Niger. The primary priority borders were those in the southern areas of the country (Nigeria, Burkina Faso, and Mali) rather than the north (Libya, Algeria). 4 Figure 1 is similar to Figure A1 in Aker (2010), but it extends the data until 2008 and adds data on road quality. 5 Since 2008, mobile phone coverage and adoption have expanded considerably in rural areas in Niger. The 2009 survey conducted by one of the authors revealed that mobile phone ownership had reached 29 percent in rural areas. 266 F I G U R E 1. Mobile Phone Subscribers, Landlines, and Road Quality, 2001–8 THE WORLD BANK ECONOMIC REVIEW Notes: Raw data obtained from Sonatel/Niger (number of landlines), Wireless Intelligence (number of mobile phone subscribers), and the World Bank (http://data.worldbank.org/country/niger). Adapted from Aker (2010). Aker and Fafchamps 267 farmers and itinerant traders. The model is based on a fair but stylized descrip- tion of how agricultural traders operate in sub-Saharan Africa (Fafchamps, Gabre-Madhin, and Minten 2005).6 Because each market operates for a limited duration on different days and markets are often far apart, it is difficult for farmers or traders to visit more than one market in a given day. In the absence of widespread communication devices, farmers and traders must visit a market to obtain price information and are limited to one per day. For Nigerien farmers, these travel costs are substantial, as the average distance to the nearest market is 7.5 km, or 1.5 hours.7 For this reason, farmers often sell their output in the village (farm-gate) or at the closest market and are reluctant to transport their output home because of transport costs. As a result, supply on any given market day is inelastic. Because each weekly market serves a large geographic area, farmers cannot coordinate supply. Consequently, supply varies randomly across market days in ways that traders cannot easily predict. If better-informed traders can select which market to visit depending upon local supply, this situation would generate a potential gain from spatial arbitrage. The introduction of mobile phone coverage in Niger reduced the cost of obtain- ing information in general and for information about local market conditions in particular. Aker (2010) estimated that the cost of obtaining price information from a market located 10 km away decreased by 35–50 percent with the introduc- tion of mobile phone coverage.8 This cost reduction had a greater effect on traders (compared with farmers) because traders were more likely to adopt mobile phones and operated in different weekly markets, many of which had mobile phone cover- age. In contrast, mobile phone coverage did not expand significantly into more remote rural villages, farmers’ primary residences, until after 2008. Our model captures these different features to illustrate how mobile phone coverage can increase spatial arbitrage and reduce price dispersion while keeping the average producer price unaffected. The Model To provide a focus, we consider a static, symmetric, one-period model without intertemporal arbitrage. Each risk-neutral trader has n markets nearby, all located at the same distance d with transport cost c.9 Each market is reachable by n traders, and each trader has a working capital of k. In the morning, the trader visits a single market, m, and purchases all possible quantities with working 6 Jensen (2007, 2010) develops a model outlining the welfare implications of costly search under exogenous supply, focusing primarily on the producers’ perspective. 7 Agricultural traders in Niger also typically relied upon personal travel to obtain price information prior to the introduction of mobile phones. 8 In 2008, a two-minute call to a market located 10 km away cost US$.50, compared to US$1 for round-trip travel using a market truck or cart. 9 This scenario encompasses two possibilities. The first is that traders and markets are placed at regular intervals on a lattice or Taurus, and traders cover partially overlapping geographical areas. The second is that n traders cover the same n producer markets and sell in the same consumer market. 268 THE WORLD BANK ECONOMIC REVIEW capital k at the going price pm. In the afternoon, the trader sells the total quantity purchased to his home market. In the morning of day t, producers bring a random quantity of agricultural goods q ~m to market m. The distribution F(q) of q ~m is the same in all markets and is known to traders, but the exact quantity in each market is unknown.10 Let E[q~m ] ¼ q  for all m. Each morning, a trader must select a market m among the n markets that he could potentially visit.11 Let d ~ m be the number of traders who happen to choose market m on a particular day t. Because traders randomize equally among all n markets, d ~ m B À1; nÀ1Á, it follows that the variance of d ~m n increases in n, the number of markets from which the traders must choose. The total demand is equivalent to d ~ m k, the number of traders multiplied by their individual working capital. The price in market m on a given day t is given by the standard supply equals demand equilibrium: pm q ~m : ~m ¼ kd ð1Þ Setting k ¼ 1 by choice of units, this reduces to pm ¼ d~ m =q ~m . In other words, the price on a given market on a given day is the same for all farmers and traders in that market, but there is spatial price variation across markets. For a given distri- bution F(q), the variance of pm is increasing in the variance of dm and hence n. The quantity that each trader i purchases on a given day is q ~m q ~i ¼ ; ð2Þ ~m d which is increasing in the number of farmers who brought their output to the market that day and decreasing in the number of traders who chose market m. In the absence of temporal arbitrage, a trader must sell in his/her home market. We assume, for simplicity, that each trader sets the sales price to cover transport costs plus a unit profit margin r. The sales price for trader i is thus ~i ¼ p p ~m þ c þ r: ð3Þ Because the profit of trader i is rq ~i , the trader would prefer to buy from markets with many farmers (i.e., a high q ~ m Þ. In ~m ) and few other traders (i.e., a low d other words, a trader could benefit from obtaining an informative signal about the realization of q ~m or d~ m . Without this signal, the trader’s best response 10 Systematic differences in F(q) across markets would generate systematic differences in average price. However, in the first approximation, this should not affect the spatial integration of prices. 11 We only consider symmetric equilibria (thus ruling out a situation whereby traders coordinate on a public randomization device), which implies that the only Nash equilibria is for each trader to randomize among each market with equal probability. Aker and Fafchamps 269 function is X n q ~ max pm pm rE½ m Š; ð4Þ 1þd^m m¼1 ^m is the number where pm is the probability that trader i will visit market m and d of traders other than trader i in market m. In a symmetric equilibrium, d ~ m is not correlated with q~m because traders do not know q ~m and all markets have the same ex ante F(q). The best response function can therefore be rewritten as X n 1 maxpm rq  pm E½ Š; ð5Þ ^m 1þd m¼1 where p^ m is the probability that the other n-1 traders visit market m. The distri- bution function for d^m is   ^m ¼ xÞ ¼ nÀ1 Prðd ^ m ÞnÀxÀ1 : ^ m Þx ð1 À p ðp ð6Þ x Given symmetry, it follows that pm ¼ p ^ m ¼ 1=n , i.e., that each trader randomiz- es equally across the n markets. Without information on q ~m or d~ m , all markets are ex ante equivalent from the point of view of traders. Informative Signals Let us now assume that trader i receives a costless ( private) informative signal s, such as a phone call, about the number of farmers q ~m and learns that E½qm jsŠ ) E½qm Š. Signal s breaks the symmetry between markets for trader i, ~ ~ whose best response function becomes " # " # q ~s X nÀ1 1 À ps q ~m maxps ps rE js þ rE ; ð7Þ ^s 1þd nÀ1 1þd^m m¼1 where ps is the probability of trader i visiting markets with a signal and pm ¼ 1 À ps =n À 1 is the probability of trader i visiting markets without a signal. If other traders do not receive the private signal, then Fðd ^s Þ ¼ Fðd ^m Þ. Hence, d^s is independent of the realized q s ; E½q ~s and s. If q ~s jsŠ and we assume that q s . q, then equation (7) can be rewritten as X nÀ1 1 À ps maxps rq  ps þ rq s C C ¼ rq  ps þ rq s C C ð1 À ps Þ; ð8Þ m¼1 nÀ1 h i ^m denotes the competition that trader i expects to face on  ; E 1=1 þ d where C 270 THE WORLD BANK ECONOMIC REVIEW market m. Solving the first-order conditions (FOC) yields the following: rq  ¼ rq s C C ; s:t: q s ¼ q ; ð9Þ which shows that if the signal is informative, playing a randomized strategy is no longer optimal. The informed trader then sets s . q ps ¼ 1 if q s , q  and ps ¼ 0 if q : ð10Þ In other words, if the signal is informative and the other i-1 traders remain unin- formed, trader i is no longer indifferent between all n markets but visits market m with probability 1 if qs . q . All else equal, the informative signal s reduces price dispersion across markets. The informed trader buys from a large surplus market, thereby increasing the price in that market, and abandons a small surplus market, thereby reducing demand and prices there. Thus, the action of the informed trader reduces the price difference between high and low surplus markets. A similar outcome is obtained if all traders receive the same signal s. To solve for the mixed strategy analytically, we assume that traders are divided into non- overlapping geographical areas consisting of one consumer market and n supply markets. We now have Fðd ^ s Þ = F ðd ^m Þ. If q  s =n ¼ q C , the signal in favor of market s is very strong, and there is a corner equilibrium: all n traders go to market s and do not visit the other markets on that day, which is the pure strategy equilibrium case. If, however, the signal is not strong, traders are not all attracted by the same market, and there is a mixed strategy equilibrium in which each trader’s best response is " # " # 1 1 maxps ps rq s E rE js þ ð1 À ps Þq ; ð11Þ ^s 1þd ^i 1þd where d^i denotes the number of traders in markets other than s. In general, ^ ^ di = dm . The condition for an interior solution is that " #, " # 1 1 s =q u;q ¼E E js ; ð12Þ ^i 1þd ^s 1þd where u represents the strength of the signal. Aker and Fafchamps 271 By symmetry, all traders face the same decision problem. Thus, their ps will all be the same:   ^s ¼ xÞ ¼ nÀ1 Prðd ðps Þx ð1 À ps ÞnÀxÀ1 ð13Þ x      ^i ¼ xÞ ¼ nÀ1 1 À ps x 1 À ps nÀxÀ1 Prðd 1À : ð14Þ x nÀ1 nÀ1 The system of three equations (12), (13), and (14) defines an implicit relationship between ps and the signal strength u. It is easy to show that u increases monotoni- cally with ps . It follows that the equilibrium ps must rise to equilibrate the FOC. In other words, @ ps/@ u . 0: a more positive signal is associated with a higher probability that traders will visit market s. This raises prices in surplus (low price) markets and lowers them in deficit (high price) markets, thereby reducing price dispersion between markets. Theoretical Predictions This model, albeit stylized, conveys the key intuition behind the empirical analy- sis. The introduction of mobile phones in Niger provided traders with access to a private signal about local supply and arguably improved the quality of such signals in terms of accuracy, detail, and timing. During our study period, the technology was primarily used by traders, with more limited access to and use by farmers. The model suggests that the introduction of mobile phone coverage should 1) lead to a shift in informed traders’ attention to market s (i.e., the surplus market with an informative signal); 2) raise producer prices in surplus (low price) markets and lower them in deficit (high price) markets; and 3) reduce spatial price dispersion between markets with mobile phone coverage. These effects need not be present for storable commodities because intertempo- ral arbitrage sets a ceiling or a floor on the price at which farmers and traders are willing to trade and can thus reduce price dispersion even in the absence of infor- mative signals.12 This implies that the above predictions apply primarily to perish- able and semi-perishable commodities for producer prices, but not necessarily to non-perishables or for consumer price markets, as is the case in Aker (2010). In this paper, we formally test the third hypothesis for non-perishable and semi-perishable commodities and provide suggestive evidence in support of the second hypothesis. 12 To illustrate our model, assume that traders receive no information about market prices, and imagine that the agricultural commodity can be stored by farmers. Consider farmers in market m who are offered a low price because few traders happen to visit market m on that day. Rather than selling at a low price, they can store and sell later, when more traders visit the market. In this case, intertemporal arbitrage will smooth prices in market m across time (Williams and Wright 1991). As a result, prices in different markets cannot diverge simply because the number of traders who visit each market varies in a stochastic manner. 272 THE WORLD BANK ECONOMIC REVIEW D ATA AND S U M M A RY S TAT I S T I C S This paper uses three primary datasets. The first is a market-level monthly panel for 37 markets over a 10-year period (1999–2008) collected by the Agricultural Market Information Service (AMIS) of Niger. This dataset includes monthly producer and consumer prices for millet, sorghum, and cowpea.13 A producer price observation is calculated as the average monthly price that farmers received for selling a given crop in that market. In addition, we have data on factors that may affect arbitrage, such as fuel prices, transport costs, rainfall, market latitude and longitude, and the dis- tance and road quality between market pairs.14 These data are combined with infor- mation on the location and date of mobile phone coverage in each market between 2001 and 2008, which was obtained from the mobile phone service providers. The second dataset is a panel survey of traders and farmers interviewed in Niger between 2005 and 2007. The survey includes traders and farmers in 32 markets and 37 villages across six geographic regions of Niger. A majority of traders in Niger are male, from the Hausa ethnic group, and have never attended school. Traders search for price information in an average of 3.9 markets and buy and sell commodities in four markets. Traders have an average of 16 years’ of trading expe- rience, and only 10 percent had changed their market since they began trading (Tack and Aker 2014). Table 1A provides summary statistics for farmers. Despite their low levels of production, on average, 25 percent of farmers sold millet, and 56 percent sold cowpea. Compared to traders, farmers traded over a smaller geograph- ic area, selling in 1.46 markets and searching for price information in 1.5 markets. Fewer than 5 percent of the villages in the farmer survey had mobile phone coverage between 2005 and 2007. As a result, inferences drawn from this dataset have low power. To provide additional insight into the relationship between mobile phone coverage and farmers’ marketing behavior, we also rely on a 2009 survey of 1,038 farm households from 100 Niger villages. The survey collected data on agricultural production and marketing behavior as well as mobile phone coverage and ownership. Table 1B provides summary statistics from this survey. Overall, many of the socio-demographic indicators are similar to those of the 2005–7 survey; households have low levels of education and are primarily from the Hausa ethnic group. Millet and cowpea are the primary crops grown by farm households in our sample, followed by sorghum and peanut. More than 70 percent of households sold cowpea, compared with 36 percent of households that sold millet. Farm households purchased and sold their staple food and cash crops in 2.3 markets and primarily sold to traders located in markets. However, 13 Because sorghum requires more rainfall than millet, it is grown in fewer geographic locations in Niger, primarily in the southern areas of the country. As a result, sorghum price data are available for fewer markets and during fewer periods and hence are subject to more missing observations than the cowpea or millet producer price data. 14 These data were obtained from the Syndicat des Transporteurs Routiers, the Direction de la Me ´o ´ te in Niger, and the trader survey. The data are the same as those used in Aker (2010), but are extended for an additional two years (from 2006 to 2008). Aker and Fafchamps 273 T A B L E 1 A . Summary Statistics Farmer Survey 2005–7 Panel A: Farmer-Level Characteristics Sample Mean Socio-Demographic Characteristics Age of respondent 49 Gender of respondent (male ¼ 0, female ¼ 1) .01 Education of respondent (0 ¼ elementary or above, 1 ¼ no education) .85 Member of Hausa ethnic group .68 Panel B. Agricultural Marketing Activities Sold millet in the past year 0.25 Sold cowpea in the past year 0.56 Purchased millet since the previous harvest 0.91 Number of hours walking to principal market 1.53 Access to a paved road .27 Number of purchase and sales markets 1.46 Member of a producers’ association 0.22 Sold to intermediary since the last harvest 0.45 Bought agricultural products on credit in the past year 0.41 Received payment in advance for harvest 0.16 Responsible for transport if selling product 0.64 Household follows market price information Personal conversations with traders and farmers 0.75 Radio (MIS) 0.09 Other 0.14 Note: Total sample size is 200 farmers across 37 villages across six regions of Niger. Respondents are primarily the household head. Source: Niger farmer survey conducted by one of the authors between 2005 and 2007. approximately 30 percent of households owned mobile phones by this time, a marked increase from the earlier survey. E M P I R I CA L S T R AT E GY We first examine the impact of mobile phone coverage on producer price disper- sion, comparing market pairs with and without mobile phone coverage using a difference-in-differences (DD) strategy similar to that of Aker (2010): Y ijk;t ¼ b0 þ b1 mobile jk;t þ X0jk;t g þ a jk þ ut þ m jk;t ; ð15Þ where Y ijk;t is the absolute value for commodity i (millet, sorghum or cowpea) of the difference in logged producer prices between markets j and k at month t.15 15 Various dependent variables have been used in the literature to measure price dispersion. The consumer search literature has used the sample variance of prices across markets over time (Pratt, Wise, and Zeckhauser 1979), the coefficient of variation (CV) across markets (Eckard 2004; Jensen 2007), or the maximum and minimum prices across markets (Pratt, Wise, and Zeckhauser 1979; Jensen 2007). The international trade literature has used the log of the price ratio between two markets or the standard deviation of price differences across markets (Engel and Rogers 1996; Parsley and Wei 2001; Ceglowski 2003). We adopt the latter approach for our core specification, but we also use the CV and the max-min as alternative specifications. 274 THE WORLD BANK ECONOMIC REVIEW T A B L E 1 B . Summary Statistics Farmer Survey 2008 Sample Mean (s.d.) Panel A: Farmer-Level CharacteristicsSocio-Demographic Characteristics Age of respondent 37.5 (12.44) Gender of respondent (male ¼ 0, female ¼ 1) .5 (.5) Education (0 ¼ No education, 1 ¼ Some education (including Coranic) .08 (.27) Member of Hausa ethnic group (1 ¼ Hausa, 0 otherwise) .72 (.45) Household size 8.37 (4.06) Household owned mobile phone in 2008 .29 (.46) Number of mobile phones owned .37 (.65) Panel B. Agricultural Marketing Activities Cultivated millet during previous harvest .99 (.06) Cultivated sorghum during previous harvest .79 (.41) Cultivated cowpea during previous harvest .95 (.23) Cultivated peanut during previous harvest .56 (.50) Sold millet since previous harvest .36 (.48) Sold sorghum since previous harvest .09 (.29) Sold cowpea since previous harvest .70 (.46) Sold peanut since previous harvest .49 (.50) Purchased millet since previous harvest .35 (.48) Purchased sorghum since previous harvest .12 (.32) Purchased cowpea since previous harvest .11 (.32) Purchased peanut since previous harvest .05 (.23) Number of purchase and sales markets for grains and cash crops 2.35 (1.26) Member of a producers’ association .38 (.49) Sold to trader in village since previous harvest .17 (.38) Sold to trader in market since previous harvest .65 (.48) Household follows market price information .75 (.43) Notes: The total sample size is 1,038 farm households across 100 villages in two regions of Niger. Respondents are either men or women within the household who are eligible for an adult ed- ucation program. Source: Data from a baseline survey collected for Project ABC in 2009 (Aker, Ksoll and Lybbert 2012). mobilejk,t is a binary variable equal to one in month t if both markets j and k have mobile phone coverage and 0 otherwise.16 The ajk 0 s are market-pair fixed effects, controlling for time-invariant factors such as geographic location, urban status, and market size. The ut 0 s are a vector of time fixed effects, either yearly or 0 monthly. We also include a set of market-pair time-varying controls (X jk,t ) likely 17 to affect spatial price dispersion, such as transport costs and drought. The 16 In all specifications, “treatment” is defined as the presence of a mobile phone tower rather than mobile phone adoption (Aker 2010). 17 A market’s urban status did not change between 1999 and 2008, so this is controlled for by including market pair and market fixed effects. Road quality in Niger was fairly stagnant during the time period under consideration. In 1995, Niger had 3,526 km of paved roads, increasing to 3,761 km in 2008, with the primary improvement occurring in 1997 ( prior to mobile phone coverage) (Figure 1). Among the markets in our sample, only 16 percent of markets received some type of road improvement between 1999 and 2008, with the majority of this improvement occurring in 2007/2008, toward the end of our sample period and well after mobile phone coverage was introduced into these markets. Aker and Fafchamps 275 parameter of primary interest is b1, with a negative value indicating that mobile phone coverage reduces price dispersion between market pairs. Following Aker (2010), we first cluster the standard errors at the market pair level. This allows for dependence over time within the market pair. As a robust- ness check, we also include market fixed effects and cluster by quarter to correct for spatial dependence across markets within a period, while allowing for some dependence between months (Aker 2010; Bertrand, Duflo, and Mullainathan 2004). Equation (15) is also estimated using an alternative specification of the depen- dent variable: Yr;t ¼ b0 þ b1 mobilepercentr;t þ X0r;t g þ ut þ ar þ mr;t ; ð16Þ where Yr,t is the difference in maximum and minimum producer prices across markets within a region during month t and mobilepercentr,t is the percentage of markets within a region that have mobile phone coverage during month t. Identification and Assumptions To interpret b1 as the causal effect of mobile phone coverage on producer price dispersion, we must assume that, conditional on all covariates, mobilejk,t is un- correlated with the error term. The DD specification controls for time-invariant unobservables, but we must also assume that there are no time-varying unob- servables that are correlated with mobile phone coverage and the outcomes of interest. We formally test the validity of these identification assumptions in several ways. First, we examine whether mobile phone coverage expanded primarily into markets where the potential for improvement in market integration was the greatest. This does not appear to be the case. Mobile phone operators cited two main determinants of tower placement: whether a location was an urban center (defined as a population of more than 35,000) and whether the location was near the southern or western borders. To verify these claims, we regress a binary vari- able for mobile phone coverage in market j at time t on j’s urban status, latitude and longitude, elevation, slope, and road quality (Buys et al. 2009; Batzilis et al. 2010). The regression results appear to confirm the criteria cited by mobile phone operators (Table 2): urban areas were more likely to receive mobile cover- age, as were markets with paved roads, a variable that is correlated with urban status (column 1). The eastern part of the country was also slightly more likely to receive mobile phone coverage earlier, as it had a higher density of border markets. Characteristics that are potentially correlated with high potential for market integration, such as elevation, slope, latitude, and market size, are not correlated with mobile phone expansion during our study period. These results are robust to the use of probit estimation (column 2). 276 THE WORLD BANK ECONOMIC REVIEW T A B L E 2 . Correlates of Mobile Phone Coverage in Niger Dependent variable: Mobile phone coverage ( ¼ 1) in market j at time t (1) (2) Log(elevation) 0.00 0.01 (0.15) (0.43) Dummy slope 0.02 0.06 (0.06) (0.17) Urban center 0.28*** 0.77*** (0.05) (0.14) Road quality 0.04 0.13 (0.05) (0.16) Latitude 2 0.01 2 0.04 (0.03) (0.09) Longitude 0.01 0.03 (0.01) (0.03) Market size 2 0.00 2 0.00 (0.00) (0.00) Constant 0.34 2 0.45 R2 0.09 0.066 Number of observations 4032 4032 Notes: Mobile phone coverage is equal to 1 in market j at time t if the market received mobile phone coverage and 0 otherwise. The slope variable is equal to 1 if the market is steeply sloped and 0 otherwise. Urban center is equal to 1 if the market has a population greater than 35,000 and 0 otherwise. Road quality is equal to 1 if the market has access to a paved road and 0 otherwise. Column 1 is OLS estimation, and column 2 is probit estimation. *** significant at the 1 percent level, ** significant at the 5 percent level, * significant at the 10 percent level. Source: Data collected from the mobile phone operators in Niger between 2001 and 2008, as well as the authors’ own market survey. Even if agricultural market performance were not an explicit rationale behind tower placement, the possibility remains that the spatial producer price disper- sion is correlated with pre-treatment time-invariant or time-varying characteris- tics that led to the placement of mobile phone towers. To determine whether this is the case, Table 3 shows the differences in means for pre-treatment (1999– 2001) outcomes and covariates at the market (Panel A) and market pair level (Panel B).18 Overall, the results suggest that there were few statistically significant differences in pre-treatment outcomes between mobile phone and non-mobile phone markets. Most differences in pre-treatment covariates are also not statisti- cally significant from zero, with the exception of a market’s urban status, thereby supporting the earlier results. Although the pre-treatment differences in producer price levels and dispersion are not significantly different from zero for millet and sorghum, there is a statistically significant difference in cowpea producer prices and dispersion. However, a statistically significant difference only exists for one 18 This table is similar to the Table 2 in Aker (2010), with the exception of the definition of the “mobile phone coverage” variable. Aker and Fafchamps 277 T A B L E 3 . Pre-Treatment Comparison of Means by Mobile Phone Coverage (1999-2001) (1) (2) No Mobile Phone Coverage Coeff Mean (s.d.) (s.e.) Panel A. Market Level Data Millet Producer Price level (CFA/kg) 93.12 6.16 (30.93) (4.09) Cowpea Producer Price level (CFA/kg) 135.19 16.56*** (36.05) (5.05) Sorghum Producer Price level (CFA/kg) 84.24 6.39 (33.27) (4.99) Drought in 1999 or 2000 0.05 0.01 (0.22) (0.03) Hausa ethnic group (Hausa ¼ 1) 0.79 2 0.19 (0.41) (0.21) Road Quality to Market (1 ¼ paved) 0.41 0.27 (0.50) (0.24) Market Size (More than 100 traders ¼ 1) 0.59 0.05 (0.50) (0.24) Distance (km) to international border 78.07 15.73 (57.78) (28.84) Urban center( . ¼ 35,000) 0 0.35*** (0.00) (0.09) Panel B. Market Pair Level Data jLn (Millet Producer price dispersion)j 0.13 0.00 (0.13) (0.01) jLn (Cowpea Producer price dispersion)j 0.17 0.03*** (0.14) (0.01) jLn (Sorghum Producer price dispersion)j 0.20 0.01 (0.05) (0.01) Distance between markets (km) 380.25 2 12.00 (240.95) (16.05) Road Quality between markets (both paved ¼ 1) 0.38 0.02 (0.49) (0.03) Transport Costs between Markets (CFA/kg) 11.02 2 0.32 (6.41) (0.42) Notes: Data from the Niger trader survey and secondary sources. In Panel A, “mobile phone” markets are those that received coverage at some point between 2001– 2008; “no mobile phone” markets are those markets that never received coverage during this period. In Panel B, “mobile phone” market pairs are pairs where both markets received mobile phone coverage at some point between 2001– 2008; “no mobile phone” market pairs are those pairs where either one or both markets never received mobile phone coverage during this period. The first column shows the mean of characteristics for “no mobile phone” markets or market pairs. Column 2 shows the regression coefficients and standard errors for the “mobile phone” markets or market pairs. All regressions include time and market or market pair fixed effects. Huber-White robust standard errors clustered by market (Panel A) and by market pair (Panel B) are in parentheses. * is significant at the 10% level, ** is significant at the 5% level, *** is significant at the 1% level. Prices are deflated by the Nigerien Consumer Price Index. Sources: Agricultural Market Information System (AMIS) (market prices), SONIDEP (fuel prices), Direction de la Meteo (drought), mobile phone operators (mobile phone coverage) and authors’ own work (market size, road quality, transport costs). 278 THE WORLD BANK ECONOMIC REVIEW of the two pre-treatment years, suggesting that this was not systematic. In addi- tion, pre-treatment cowpea price dispersion is lower in non-mobile phone markets. Hence, if our findings are biased because of the non-random placement of phone towers, this bias is most likely in the direction of underestimating the effect of mobile phone coverage. The key identification assumption of equation (15) is that of parallel trends across mobile phone and non-mobile phone markets. This assumption might be violated if we do not control for time-varying characteristics, such as road quality, that are simultaneously correlated with mobile phone coverage and price dispersion. We therefore conduct a falsification test by estimating equation (15) using data from before the introduction of mobile phones (Table 4). The ratio- nale behind this test is that if mobile phone and non-mobile phone markets follow different time trends, this difference should have already been apparent before mobile phone coverage was introduced (Aker 2010). We find that the pre- intervention trends for the log of cowpea producer prices (column 3) and cowpea and sorghum producer price dispersion (columns 4 and 6) are not significantly different from zero for markets and market pairs that ever received mobile phone coverage compared with those that did not. However, there seem to be somewhat differing trends for millet and sorghum producer prices and millet producer price dispersion. This finding raises some concerns regarding the parallel trend as- sumption, especially for millet producer price dispersion. Nevertheless, mobile phone markets had relatively higher producer price dispersion prior to treatment, suggesting that our findings may underestimate the effects on millet producer price dispersion. In addition, these differences are present during one of the pre-treatment years rather than both, suggesting that the differences are not systematic.19 RE S U LTS We first present the impacts of mobile phone coverage on producer price disper- sion for different commodities. Because these results might differ by market char- acteristics, we then present heterogeneous treatment effects by distance, road quality, period of year, and market type. We end with the results on the impact of mobile phone coverage on producer-consumer margins and producer price levels. 19 These tests do not test for other time-varying unobserved factors that might be simultaneously correlated with mobile phone coverage and the outcomes of interest, such as road quality, landline coverage, donor funding, and private sector investment. Some of these factors, such as changing relations with France in the north and an increase in anti-terrorism activity, either occurred outside of our study sample (e.g., in the Saharan desert) or outside of our study period. Other changes occurring in Niger over this time (such as the introduction of Chinese investment into Niger or increased aid in the wake of the 2005 food crisis) would need to be correlated with an increase in mobile phone rollout during our time period. Although Chinese investment could be a potentially important confounding factor, the growth of this investment occurred after most of the markets in our sample already had mobile phone coverage. T A B L E 4 . Pre-Treatment Producer Price Trends by Mobile Phone Coverage Millet Cowpea Sorghum Commodity (1) (2) (3) (4) (5) (6) Dependent Variable ln(Pit) jln(Pit)-ln(Pjt)j ln(Pit) jln(Pit)-ln(Pjt)j ln(Pit) jln(Pit)-ln(Pjt)j Mobile phone market*time 2 0.08** 0.02* 2 0.01 2 0.02 2 0.10** 0.02 (0.03) (0.01) (0.04) (0.02) (0.04) (0.02) Time ( ¼ 1 if 2000/2001, 0 if 1999/2000) 0.56*** 2 0.04*** 0.09* 2 0.09*** 0.20*** 2 0.07*** (0.04) (0.00) (0.04) (0.02) (0.05) (0.02) Market fixed effects Yes Yes Yes Yes Yes Yes Market pair fixed effects No Yes No Yes No Yes Additional covariates Yes Yes Yes Yes Yes Yes R2 0.86 0.08 0.73 0.12 0.88 0.1 Number of observations 423 7,190 408 6,696 302 3,718 Notes: Huber-White robust standard errors clustered by market (columns 1 and 3) or market pair (columns 2 and 4) are in parentheses. * significant at Aker and Fafchamps the 10 percent level, ** significant at the 5 percent level, *** significant at the 1 percent level. All prices are in 2001 CFA. Source: Agricultural Market Information System (AMIS) (market prices), SONIDEP (fuel prices), Direction de la Meteo (drought), mobile phone opera- tors (mobile phone coverage) and authors’ own work (market size, road quality, transport costs). 279 280 THE WORLD BANK ECONOMIC REVIEW Average Effects of Mobile Phone Coverage Table 5 presents the regression results of equation (15) for cowpea (columns 1– 4), millet (columns 5–8), and sorghum (columns 9–12). According to our model, we expect mobile phone coverage to reduce price dispersion for cowpea more than for millet or sorghum, as cowpea is a semi-perishable commodity. Controlling for yearly, monthly, and market pair fixed effects (column 1), we find that mobile phone coverage reduces cowpea producer price dispersion by 6.3 percent, with a statistically significant effect at the 1 percent level. This result is robust to the introduction of additional covariates that also affect producer price dispersion across markets (column 2), such as drought and transport costs. The result is also similar when including market fixed effects with the standard errors clustered by quarter (column 3). Similar to Aker (2010), we also include a binary variable equal to one when only one market in a pair has mobile phone coverage (column 4). Using the most conservative estimate of all of the specifications, the in- troduction of mobile phones is associated with a 6 percent reduction in cowpea producer price dispersion compared to market pairs without mobile phones.20 Columns 5 to 12 contain similar regressions for millet and sorghum, two less perishable commodities. Mobile phone coverage reduces millet producer price dispersion across markets by as little as 0.1 percent without a statistically signifi- cant effect. The magnitude and statistical significance of this effect is similar across all specifications (columns 5 –8) and is similar for sorghum (columns 9–12). Taken together, these results are consistent with the idea that even limited intertemporal storage may act as a buffer on local market producer price fluctuations, for less perishable commodities. A concern with the producer price data used in these estimations is the number of missing observations. Because the demand for staple grains is relative- ly constant throughout the year, consumer price data are readily available for each market and each month (Aker 2010). In contrast, farmers in Niger do not have sufficient stocks to sell throughout the year. As a result, producer price data are not available for some markets during certain periods of the year, especially during the season immediately prior to the annual harvest. To check the robustness of our results to potential selection bias generated by missing data, we re-estimate equation (15) in two different ways. We first use a two- stage Heckman procedure, estimating a selection equation on market-pair data and adding the resulting inverse Mills’ ratio as a separate regressor to equation (15) (Heckman 1979). These results are shown in the supplemental appendix (tables S1 and S2, available at http://wber.oxfordjournals.org/). The results for cowpea are similar in magnitude and statistical significance to those reported in Table 5. We also re-estimate equation (15) using a balanced panel of market pairs that have a full set of price data for all time periods in the sample (table S2, columns 5–8). The 20 Aker (2010) also included cross-border markets in the specification. This is not possible for producer price dispersion, as producer price data are not available from cross-border markets. Thus, all of the regressions are only for markets within Niger. T A B L E 5 . Impact of Mobile Phone Coverage on Producer Price Dispersion Cowpea Millet Sorghum Dependent variable: jln(Pit-)-ln(Pjt)j (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) Mobile coverage both markets 2 0.06*** 2 0.06*** 2 0.06*** 2 0.08*** 0.00 0.00 0.00 0.00 0.00 2 0.00 0.01 0.01 (0.01) (0.01) (0.01) (0.01) (0.00) (0.00) (0.01) (0.00) (0.01) (0.01) (0.01) (0.02) Mobile coverage one market 2 0.01*** 2 0.00 2 0.00 (0.00) (0.00) (0.01) Other covariates No Yes Yes Yes No Yes Yes Yes No Yes Yes Yes Market pair fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Year fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Monthly fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Market fixed effects No No Yes Yes No No Yes Yes No No Yes Yes Market pair-specific time trend Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Number of observations 39,120 39,120 39,120 39,120 39,002 39,002 39,002 39,002 13,280 13,280 13,280 13,280 R-squared 0.154 0.165 0.37 0.38 0.09 0.09 0.32 0.32 0.10 0.10 0.20 0.20 Notes: For market pairs, mobile phone coverage ¼ 1 in period t when both markets have mobile phone coverage and 0 otherwise. Additional covariates Aker and Fafchamps include inter-market transport costs at time t and the presence of drought in both markets at time t. Huber-White robust standard errors clustered by market pair are in parentheses in columns 1, 2, 4 and 5, 6, and 8. Huber-White robust standard errors clustered at the quarterly level are also provided in columns 3 and 7. All prices are deflated by the Nigerien Consumer Price Index (CPI). *** significant at the 1 percent level, ** significant at the 5 percent level, * signifi- cant at the 10 percent level. Sources: Agricultural Market Information System (AMIS) (market prices), SONIDEP (fuel prices), Direction de la Meteo (drought), mobile phone opera- tors (mobile phone coverage) and authors’ own work (market size, road quality, transport costs). 281 282 THE WORLD BANK ECONOMIC REVIEW point estimates are slightly smaller, but the coefficient estimates remain significant at the 1 percent level. It is important to note that our results for millet producer price dispersion differ from the results of Aker (2010), who found that mobile phone coverage reduces consumer price dispersion for millet by 10 percent (Aker 2010). Both findings are consistent if millet is stored primarily in production areas. When there are insufficient stocks in consumer markets, unanticipated demand shocks are a source of price fluctuation. The effect of shocks on consumer prices can only be smoothed if traders can quickly determine where to send more supplies. Using the theoretical model in reverse, we see that by facilitating the efficient al- location of millet from producer to consumer markets, mobile phones can explain the reduction in price dispersion across consumer markets. Table 6 presents results based on equation (16) using the max-min producer price spread (in CFA/kg) of markets across a region as the dependent variable (Jensen 2007). The key independent variable of interest is the intensity of mobile phone coverage within a region (mobilepercentr,t) rather than the coverage status of a particular market or market pair. Controlling for year and region fixed effects, we find that an increase in the density of mobile phone coverage within a region leads to a reduction of 26 CFA/kg in the max-min price spread of cowpea producer prices (column 1), with a statistically significant effect at the 10 percent level. There is no statistically significant impact on the max-min producer price spread for millet (column 2) or sorghum (column 3). Although the results in Table 6 are not directly comparable to those in Table 5, they demonstrate that our findings are not an artifact of the dyadic specification. T A B L E 6 . Impact of Mobile Phone Coverage on the Max-Min Price Spread (CFA/kg) Cowpea Millet Sorghum Dependent variable: Max-Min Price Spread (CFA/kg) within a Region (1) (2) (3) Percentage of markets with mobile phone 2 26.65* 2 0.27 17.33 coverage in region j at month t (9.95) (2.88) (12.28) Additional covariates No No No Year fixed effects Yes Yes Yes Monthly fixed effects Yes Yes Yes Number of observations 3107 3029 2094 R-squared 0.42 0.36 0.460 Notes: The max-min price spread is the difference between the maximum and minimum produc- er price for commodity i among markets in a given region at time t. The coefficient of variation is the standard deviation of producer prices among markets in a region a time t divided by the mean of producer prices for markets in a region at time t. Huber-White robust standard errors clustered at the regional level are in parentheses. *** denotes significant at 1 percent level, ** denotes significant at .05 percent level and * denotes significant at .10 percent level. Source: Agricultural Market Information System (AMIS) (market prices), SONIDEP (fuel prices), Direction de la Meteo (drought), mobile phone operators (mobile phone coverage) and authors’ own work (market size, road quality, transport costs). Aker and Fafchamps 283 Heterogeneous Effects of Mobile Phone Coverage From our theoretical model, the effect of mobile phone coverage on producer price dispersion should be larger for markets and time periods in which access to information reduces the spatial misallocation of traders across markets. We expect coordination failures among traders to be strongest when search costs are high, i.e., when markets are distant and transport costs are large because of poor quality roads. In addition, we expect trader miscoordination to be highest at times of the year when markets are thin (i.e., outside of the harvest period). Heterogeneous effects by market type, whether surplus or deficit, depend on the commodity. For semi-perishable crops, both deficit and surplus markets experi- ence shocks that cannot be smoothed by storage and thus benefit from better spatial arbitrage. For less perishable crops such as millet and sorghum, inter- temporal arbitrage reduces the potential benefits from spatial arbitrage. Because grain storage in Niger is undertaken predominantly by market agents near surplus markets, we would therefore expect mobile phone coverage to reduce millet and sorghum price dispersion primarily in deficit markets, with ambiguous predictions for cowpea. Table 7 estimates the heterogeneous effects of mobile phone coverage on pro- ducer price dispersion for cowpea.21 The regression specifications are similar to those in Table 5, except for the inclusion of the interaction term and the sub- group of interest. Column 1 looks at heterogeneous effects by distance, a binary variable that is equal to one if markets are more than 350 km apart. Consistent with the theoretical predictions, the interaction term is negative and statistically significant at the 1 percent level, implying that mobile phone coverage reduces price dispersion by 7 percent for markets located more than 350 km apart, com- pared to a 5 percent reduction for markets in closer proximity. Column 2 in- cludes an interaction term between mobile phone coverage and a binary variable for paved roads. Although the coefficient is consistent with the theoretical predic- tions, this effect is not statistically significant at conventional levels. Looking at the impact by period of year (column 3), the coefficient on the interaction term is positive and statistically significant, suggesting that mobile phone coverage reduces producer price dispersion less during the harvest period compared with other seasons, again in agreement with theoretical predictions. Finally, columns 4 and 5 introduce an interaction term between mobile phone coverage and market type (“surplus” market), which is defined as a market with surplus pro- duction.22 In column 4, the surplus variable is equal to one if both markets in a 21 We also conducted the heterogeneity analysis by the market’s landline status prior to mobile phone coverage and found no statistically significant effects. The results are available upon request. 22 The Niger Agricultural Market Information System (AMIS) defines four different types of markets: producer, consumer, wholesale, and border. These categories are not mutually exclusive and are open to interpretation. Here, we regard as surplus markets those that are primarily classified as producer markets (i.e., those markets that are located in surplus regions and serve as major trading points for farmers to sell their produce). 284 THE WORLD BANK ECONOMIC REVIEW T A B L E 7 . Heterogeneous Impact of Mobile Phones on Producer Price Dispersion Cowpea Dependent variable: jln(Pit-)-ln(Pjt)j (1) (2) (3) (4) (5) Mobile coverage both 2 0.05*** 2 0.07*** 2 0.07*** 2 0.06*** 2 0.06*** markets (0.01) (0.01) (0.01) (0.01) (0.01) Mobile coverage*distance 2 0.02*** (distance ¼ 1 (0.01) if . 350 km) Mobile coverage*road 0.01 quality (Paved ¼ 1) (0.01) Mobile coverage*harvest 0.02** (Harvest ¼ 1) (0.01) Mobile coverage*surplus 2 0.01** market (Both markets (0.00) are surplus ¼ 1) Mobile coverage*surplus 2 0.01 market (One market is (0.00) surplus ¼ 1) Joint effect significant Yes Yes Yes Yes Yes Other covariates Yes Yes Yes Yes Yes Market pair fixed effects Yes Yes Yes Yes Yes Year fixed effects Yes Yes Yes Yes Yes Monthly fixed effects Yes Yes Yes Yes Yes Number of observations 38,820 38,820 38,820 38,820 38,820 R-squared 0.38 0.37 0.37 0.37 0.37 Notes: Each column is a separate regression. For market pairs, mobile phone coverage ¼ 1 in period t when both markets have mobile phone coverage and 0 otherwise. Additional covariates include CFA/kg inter-market transport costs at time t and the presence of drought in one market. Huber-White robust standard errors clustered by market pair are in parentheses. All prices are de- flated by the Nigerien Consumer Price Index (CPI). Includes data for market pairs within 900 km of each other. *** significant at the 1 percent level, ** significant at the 5 percent level, * significant at the 10 percent level. Sources: Agricultural Market Information System (AMIS) (market prices), SONIDEP (fuel prices), Direction de la Meteo (drought), mobile phone operators (mobile phone coverage) and authors’ own work (market size, road quality, transport costs). pair are surplus markets and 0 otherwise. In column 5, the variable is equal to one if one market is a surplus market and the other is a deficit market. Both coef- ficient estimates are negative, suggesting that the reduction in price dispersion is stronger in surplus markets compared with deficit markets. The magnitude of the effect is small, however, and is not statistically significant in column 5. We find similar results for millet and sorghum (table S3). Effects on Gross Trade Margins and Producer Price Levels The previous results show that the introduction of mobile phone coverage reduced cowpea producer price dispersion, with stronger effects for markets Aker and Fafchamps 285 where there was potential miscoordination among traders. With the introduction of mobile phone coverage, improved information flows are expected to reduce traders’ costs. With sufficient competition among traders, reduced trade costs should reduce the average gross trade margin, that is, the difference between average consumer and producer prices in a market.23 As a result, consumer prices should fall on average, and/or producer price levels should rise. This effect should thus only affect price differences between geographically distinct surplus and deficit markets, especially if inter-temporal arbitrage is not feasible. To investigate whether mobile phone coverage affected gross trade margins, we estimate equation (15) using the absolute value of the difference in logged consumer prices in market j (a deficit market) and logged producer prices in market k (a surplus market). The sample thus only includes market pairs with a deficit and surplus market. In this specification, b1 measures the reduction in the gross trade margins associated with the introduction of mobile phone coverage.24 In order to understand how mobile phone coverage affected the components of the gross margin, we also estimate the following equation: Y ij;t ¼ b0 þ b1 mobile j;t þ X0j;t g þ aj þ ut þ m j;t ; ð17Þ with two dependent variables: (1) the log of producer prices in surplus market j in month t and (2) the log of consumer prices in deficit market j in month t. In this specification, mobilej,t is a binary variable equal to one at month t if market j has mobile phone coverage and 0 otherwise. Xj,t is a vector of control variables thought to affect producer price levels on market j, such as the occurrence of drought. The u 0j s are market fixed effects, controlling for geographic location, urban status, and market size, and ut are time fixed effects (either monthly or yearly) that control for time-varying aggregate factors. Standard errors are clus- tered at the market level. We also use the intra-annual price coefficient of varia- tion of commodity i on market j at year t as an alternative measure of producer welfare to producer prices. Table 8 presents the regression results for gross trade margins and equation (17) for cowpea (columns 1–4) and millet (columns 5–8). We find no 23 For the gross trade margin to change, consumer prices should fall, producer prices should rise, or both. This relative price change is akin to a standard tax incidence question: if the short-run price elasticity of demand is larger than the short-run price elasticity of supply, the average consumer price will fall by more than the average producer price rises. The demand for staple food is, in all likelihood, price inelastic. In contrast, supply may be more price elastic if farmers store their output or have alternative uses (e.g., cowpea cakes). Without independent evidence on short-run price elasticities in Niger, we are unable to make strong predictions either way. However, we expect no systematic effect on producer and consumer price differences within the same market because mobile phone coverage is unlikely to affect the spatial allocation of trade within a single market. 24 To answer this question precisely, we could use trader-level data on gross margins at different levels of the supply chain. Although we have trader-level data over a two-year period, we do not have these data over the full period of our sample. 286 THE WORLD BANK ECONOMIC REVIEW statistically significant effects of mobile phone coverage across markets on gross trade margins for cowpea (column 1), millet (column 5), or sorghum (not shown).25 One possibility is that the efficiency gains (i.e., economizing on search costs) were not large enough to yield a statistically significant reduction in gross margins. It is also conceivable that the net margins improved for traders but that this improvement did not occur during the study period. Turning to the regression results for equation (17), we find no effect of mobile phone coverage on producer price levels in surplus markets for cowpea (column 2), millet (column 6), or sorghum (not shown). For deficit markets, we also do not find any statistically significant effects of mobile phone coverage on consumer prices for any of the commodities.26 These findings are in line with our earlier findings that gross margins did not fall, even for cowpeas. They are also consistent with our model, which predicts a reduction in the spatial vari- ance of prices but not necessarily an effect on average prices across surplus or deficit areas.27 Using the intra-annual coefficient of producer price variation as the dependent variable, we find that mobile phone coverage reduces the average intra-annual coefficient of variation by 6 percent for cowpea, with a statistically significant effect at the 1 percent level (column 4). Given that the pre-treatment intra-annual coefficient of variation of cowpea price is 26 percent, this represents a significant reduction in the intra-annual price risk for cowpea farmers in Niger. We find neg- ative but not statistically significant effects for millet (column 8) and sorghum (not shown). These findings are also in line with the model predictions in that better spatial arbitrage reduces week-to-week price variation, but only when this role cannot be assumed by inter-temporal arbitrage. A limitation of the market-level data is that mobile phone coverage at the village level was still relatively limited by 2008. Thus, we might be underpowered to estimate effects on producer price levels. To gain more insight into this issue, we use farm household survey data collected in two regions by one of the authors in 2009, by which time mobile phone coverage had begun to reach more rural areas. Across the 100 villages in our sample, all had mobile phone coverage by the end of 2009, and 30 percent of the farmers in these villages owned mobile phones. Although there are obvious limitations to using this cross-sectional survey, the data complement the market-level data. 25 The results in columns 1 and 5 exclude cross-border markets (where consumer price data are available). 26 Restricting the sample to consumer price data during 1999–2007, we find that millet consumer prices decrease in deficit markets by 1.3-2.8 percent, with a statistically significant effect. The millet results are consistent with Aker (2008). 27 This categorization is similarly true for a market’s status as a collection, retail, or wholesale market, which is correlated with its categorization as surplus or deficit. It is also conceivable that our categorization of surplus and deficit markets is too imprecise. A market’s “surplus” or “deficit” status may vary across seasons and years. In the absence of time-varying data on trade flows, we can only categorize markets based on the average trading patterns, which might be an oversimplification. T A B L E 8 . Impact of Mobile Phone Coverage on Gross Margins and Prices Cowpea Millet Dependent variable: ln(PCit)-ln(PPit) ln(Pit) ln(Cit) Intra-annual CV ln(PCit)-ln(PPit) ln(Pit) ln(Cit) Intra-annual CV (1) (2) (3) (4) (5) (6) (7) (8) Mobile phone coverage 0.01 0.00 0.02 2 0.06*** 2 0.00 2 0.00 2 0.005 2 0.02 (0.01) (0.02) (0.013) (0.02) (0.01) (0.01) (0.008) (0.02) Other covariates Yes Yes Yes No Yes Yes Yes No Market pair fixed effects Yes No No Yes Yes No No Yes Market fixed effects No Yes Yes Yes No Yes Yes Yes Year fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Monthly fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Market-specific time trends No Yes Yes Yes No Yes Yes Yes Cross-border markets No No No No Yes No No No Number of observations 28,035 1,193 1,861 3,033 27,958 1,153 1,875 2,958 R-squared 0.52 0.84 0.74 0.44 0.60 0.89 0.86 0.55 Notes: For market pairs, mobile phone coverage ¼ 1 in period t when both markets have mobile phone coverage and 0 otherwise. Columns 1, 5, and 9 are the difference in log consumer prices in deficit markets and producer prices in surplus markets. Columns 2, 6, and 10 are the log of producer prices in Aker and Fafchamps surplus markets, whereas columns 3, 7, and 11 are the log of consumer prices in deficit markets. Additional covariates include the presence of drought in both markets at time t. Huber-White robust standard errors clustered by market pair are in parentheses. All prices are deflated by the Nigerien Consumer Price Index (CPI). *** significant at the 1 percent level, ** significant at the 5 percent level, * significant at the 10 percent level. Source: Agricultural Market Information System (AMIS) (market prices), SONIDEP (fuel prices), Direction de la Meteo (drought), mobile phone opera- tors (mobile phone coverage) and authors’ own work (market size, road quality, transport costs). 287 288 THE WORLD BANK ECONOMIC REVIEW Our dependent variable of interest is the log of producer prices received for a variety of commodities, namely millet, sorghum, cowpea, peanut, sesame, onion, calabash, and okra. The estimating equation is: Y ij ¼ b0 þ b1 mobile own j þ X0j g þ ak þ m jk ; ð18Þ where Yj is the log price of commodity i received by farmer j, mobileown is a binary variable for whether the household owns a mobile phone, X j0 is a vector of farmer-specific covariates to control for factors that potentially affect mobile phone ownership and prices received (such as land ownership, asset ownership, gender, and ethnicity), and a is a set of village fixed effects. Farmers with a phone may access price information more easily, but this will not be reflected in a higher producer price unless farmers can use this information for arbitrage. b1 may be biased upward if mobile phone ownership is correlated with unobserved factors, such as motivation or intrinsic marketing ability. Table 9 shows the results for equation (18). For all commodities, Mobile phone ownership is not associated with a significantly higher price received (Panel A).28 These findings are similar to Fafchamps and Minten (2012), who show that a mobile phone-based price information intervention in India is not associated with a higher producer price levels.29 Mechanisms To understand the mechanisms behind our market-level results, we look for direct evidence of information acquisition and arbitraging behavior among dif- ferent market agents. Our model suggests that mobile phone coverage improved traders’ access to information on producer prices in different markets, thereby fa- cilitating spatial arbitrage. At the same time, mobile phone coverage may improve farmers’ access to information, thereby allowing them to engage in spatial arbitrage. Tack and Aker (2014) measure the impact of mobile phone coverage on traders’ behavior, primarily focusing on their search behavior. Using the trader- level dataset, they find that the duration of mobile phone coverage is associated with an increase in traders’ numbers of search markets and the number of market contacts, with statistically significant effects. Although they do not provide direct evidence of a change in traders’ arbitrage behavior, the results do not appear to be driven by traders’ selection into mobile phone markets, changes in the composition of traders, or increased collusion among traders. 28 There are only 411 observations in Panel A (Table 9) because the question was only asked of farmers who had sold the relevant commodities since the previous harvest (and thus could report a producer price). We did not impute the missing values with a zero price. 29 Our results differ from Goyal (2010), who finds that internet kiosks increase producers’ soybean prices in India. The technology is different from mobile phone coverage because it provides both price and quality information to farmers. Aker and Fafchamps 289 T A B L E 9 . Correlation between Mobile Phone Ownership, Producer Prices, and Farmers’ Marketing Behavior (1) (2) Mean of Non-Mobile Phone Households Coeff (s.e.) Panel A: Producer Prices ln(Producer Price Millet) 158 2 0.09 (51) (0.12) ln(Producer Price Sorghum) 139 2 0.17 (55) (0.16) ln(Producer Price Cowpea) 207 0.01 (139) (0.04) ln(Producer Price Peanut) 117 0.07 (53) (0.05) ln(Producer Price Millet-Market Price Millet) 2 0.12 2 0.11 (0.46) (0.13) ln(Producer Price Millet-Market Price Cowpea) 2 0.23 0.06 (0.52) (0.09) ln(Producer Price Millet-Market Price Peanut) 2 166 2 114.09 (414) (213.81) Number of observations 411 Panel B: Farmer Marketing Behavior Household follows price information 0.73 0.07* (0.44) (0.03) Price information from traders market useful 0.66 2 0.10*** (0.47) (0.04) Price information from mobile phones useful 0.12 0.12*** (0.32) (0.03) Price information from friends useful 0.71 2 0.08* (0.45) (0.05) Number of purchase and sales markets 2.3 0.14 (1.17) (0.12) Number of observations 811 Notes: The total sample size is 1,038 farm households across 100 villages in two regions of Niger. Respondents are either men or women within the household who are eligible for an adult ed- ucation program. Each row represents a separate regression, controlling for household mobile phone ownership, ethnicity, gender, and village-level fixed effects. Huber-White robust standard errors clustered at the village level are in parentheses. * significant at the 10 percent level, ** signifi- cant at the 5 percent level, *** significant at the 1 percent level. Source: Data from a baseline survey collected for Project ABC in 2009 (Aker, Ksoll and Lybbert 2012). Table 9 (Panel B) provides some suggestive evidence of the impact of mobile phone coverage on farmers’ behavior using a specification similar to equation (18). Although the coefficient estimates are biased, we find that mobile phone 290 THE WORLD BANK ECONOMIC REVIEW ownership is associated with an increase in farmers’ probability of searching for price information and that mobile phones became a more useful source of such information. Unlike the trader results, farmers with mobile phones do not in- crease the number of search markets or the number of sales and purchase markets.30 This finding suggests that despite increased access to information, farmers in Niger did not change their marketing behavior. CONCLUSION This paper provides some estimates of the nature, magnitude, and distribution of the effects of mobile phones on producer market performance in Niger. Although mobile phone coverage did not reach remote rural areas during the period of our sample, it reduced the spatial dispersion of producer prices for cowpea, a semi-perishable commodity. It did not affect spatial price dispersion for millet or sorghum, two storable grains. We also find a stronger reduction in producer price dispersion in remote markets and during periods when markets were thin. The reduction in price dispersion did not increase average producer prices, but it did reduce the intra-annual price variation for cowpea. This paper provides additional empirical evidence of the importance of infor- mative signals for market efficiency and the differential impacts by crop. Combined with the results in Aker (2010) and Tack and Aker (2014), our results indicate that the introduction of mobile phones generated net efficiency gains in agricultural markets in Niger. However, we find no evidence suggesting that these gains translate into higher average prices for the primary suppliers of these commodities, rural farmers. These findings are central to the current debate on the role of information technology in promoting economic development. Information technology is often considered a low development priority. However, some believe that by re- ducing communication costs over long distances, mobile phones can reduce poverty among rural households. The results presented here indicate that the impact of technology can differ substantially by the type of crop, the type of market, and the time of year, even within the same country. Differences in impact can be linked to differences in arbitrage opportunities and market behav- ior for these crops and between agents. REFERENCES Aker, J. C. 2008. “Does Digital Divide or Provide? Information Technology, Search Costs and Cereal Market Performance in Niger.” Working Paper 177. Bureau for Research and Economic Analysis of Development (BREAD), Duke University, Durham, NC. 30 Aker and Ksoll (2013) report similar findings: a mobile phone-based education intervention (which was randomized at the village level) occurring between 2009 and 2011 increased farmers’ access to price information but did not change farmers’ marketing behavior or the farm-gate price received. Aker and Fafchamps 291 Aker, J. C. 2010. “Information from Markets Near and Far: The Impact of Mobile Phones on Agricultural Markets in Niger.” American Economic Journal: Applied Economics 2 (3) : 46– 59. Aker, J. C., and C. Ksoll. 2013. “Information Technology and Farm Households in Niger.” Unpublished mimeo. Aker, J. C., C. Ksoll, and T. J. Lybbert. 2012. “Can Mobile Phones Improve Learning? Evidence from a Field Experiment in Niger.” American Economic Journal: Applied Economics 4 (4) : 94– 120. Aker, J. C., and I. M. Mbiti. 2010. “Mobile Phones and Economic Development in Africa.” Journal of Economic Perspectives 24 (3) : 207–32. Anderson, M. L., and J. Magruder. 2012. “Learning from the Crowd: Regression Discontinuity Estimates of the Effects of an Online Review Database.” The Economic Journal 122 (563) : 957 –89. Autor, D. H. 2001. “Wiring The Labor Market.” Journal of Economic Perspectives 15 (1) : 25–40. Batzilis, D., T. Dinkelman, E. Oster, R. Thornton, and D. Zanera. 2010. “New Cellular Networks in Malawi: Correlates of Service Rollout and Network Performance.” NBER Working Paper 16616. World Bank, Washington, DC. Bertrand, M., E. Duflo, and S. Mullainathan. 2004. “How Much Should We Trust Differences- in-Differences Estimates?” Quarterly Journal of Economics 119 (1) : 249 –75. Brown, J. R., and A. Goolsbee. 2002. “Does the Internet Make Markets More Competitive? Evidence from the Life Insurance Industry.” Journal of Political Economy 110 (3) : 481– 507. Buys, P., S. Dasgupta, T. S. Thomas, and D. Wheeler. 2009. “Determinants of a Digital Divide in Sub-Saharan Africa: A Spatial Econometric Analysis of Cell Phone Coverage. World Development 37 (9) : 1494–505. Ceglowski, J. 2003. “The Law of One Price: Intranational Evidence for Canada.” Canadian Journal of Economics 36 (2) : 373–400. Engel, C., and J. H. Rogers. 1996. “How Wide is the Border?” American Economic Review 86 (5) : 1112–25. Eckard, E. W. 2004. “The ‘Law of One Price’ in 1901.” Economic Inquiry 42 (1) : 101–10. Fafchamps, M., E. Gabre-Madhin, and B. Minten. 2005. “Increasing Returns and Market Efficiency in Agricultural Trade.” Journal of Development Economics 78 (2) : 406–42 Fafchamps, M., and B. Minten. 2012. “Impact of SMS-Based Agricultural Information on Indian Farmers.” World Bank Economic Review 26 (3) : 383–414. Goyal, A. 2010. “Information, Direct Access to Farmers and Rural Market Performance in Central India.” American Economic Journal: Applied Economics 2 (3) : 22– 45. Heckman, J. J. 1979. “Sample Selection Bias as a Specification Error.” Econometrica 47 (1) : 153– 62. Jackai, L. E. N., and R. A. Daoust. 1986. “Insect Pests of Cowpeas.” Annual Review of Entomology 31 (1) : 95– 119. Jensen, R. 2007. “The Digital Provide: Information (Technology), Market Performance and Welfare in the South Indian Fisheries Sector.” Quarterly Journal of Economics 122 (3) : 879–924. ———. 2010. “Information, Efficiency and Welfare in Agricultural Markets.” Agricultural Economics 41 (S1) : 203– 16. Murdock, L. L., R. E. Shade, L. W. Kitch, G. Ntoukam, J. Lowenberg-DeBoer, and J. E. Huesing 1997. “Post-harvest Storage of Cowpea in sub-Saharan Africa.” In B. B. Singh, D. R. Mohan-Raj, K. E. DashiellL. E. N. Jackai, eds., Advances in Cowpea Research International Institute of Tropical Agriculture. Ibadan, Nigeria: JIRCAS. Muto, M., and T. Yamano. 2009. “The Impact of Mobile Phone Coverage Expansion on Market Participation: Panel Data Evidence from Uganda.” World Development 37 (12) : 1887–96. Nakasone, E. 2013. “The Role of Price Information in Agricultural Markets: Experimental Evidence from Rural Peru.” Paper presented at the 2013 Annual Meeting of the Agricultural and Applied Economics Association, August 4 –6. AAEA: Washington, DC. Parsley, D. C., and S. Wei. 2001. “Explaining the Border Effect: the Role of Exchange Rate Variability, Shipping Costs, and Geography.” Journal of International Economics 55 (1) : 87– 105. 292 THE WORLD BANK ECONOMIC REVIEW Pratt, J. W., D. A. Wise, and R. Zeckhauser. 1979. “Price Differences in Almost Competitive Markets.” The Quarterly Journal of Economics 93 (2) : 189– 211. Reinganum, J. F. 1979. “A Simple Model of Equilibrium Price Dispersion.” Journal of Political Economy 87 (4) : 851– 8. Rosenbaum, P. R. 2002. Observational Studies, 2nd edition. New York: Springer. Stahl, D. O. 1989. “Oligopolistic Pricing with Sequential Consumer Search.” American Economic Review 79 (4) : 700– 12. Stigler, G. J. 1961 “The Economics of Information.” Journal of Political Economy 69 (3) : 213– 25. Tack, J. B., and J. C. Aker. 2014. “Information and Firms’ Search Behavior.” Economics Department, Tufts University. Unpublished mimeo. United Nations Development Program. 2013. Human Development Report 2013: The Rise of the South: Human Progress in a Diverse World. New York: UNDP. Williams, J., and B. Wright. 1991. Storage and Commodity Markets. Cambridge, UK: Cambridge University Press, 1991. World Bank. 2006. Annual Progress Reports of the Poverty Reduction Strategy Paper Joint Staff Advisory Note Niger. Joint Staff Advisory Note 38271-NE. World Bank, Poverty Reduction and Equity Department Washington, DC. ———. 2010. Improving Education and Developing Skills for Economic Growth in Niger. Education Study Report 69946. World Bank, Economic and Sector Work Department, Washington, DC. ———. 2013. Country Partnership Strategy for the Republic of Niger for the Period FY13-FY16. World Bank, Washington, DC.